Nvidia's Les Karpas Says Robotics Is Still Hunting Its ChatGPT Moment

Nvidia's Les Karpas is set to take the stage at TechCrunch Disrupt 2026 to address a question that has dogged the robotics sector for years: why machines that walk, grasp and lift have not crossed into everyday use the way chatbots did. Karpas, who works inside the company whose chips underpin most modern AI training, will argue that the gap is not about hardware alone. His session lands as the conference pushes early registration, with passes discounted by up to $200 for those who sign up before September 25.

The comparison to ChatGPT is not idle. When large language models hit a usable threshold, deployment spread through software updates and cloud APIs, and costs fell fast because the underlying compute was already in place. Robots face a different problem: every unit has to be built, shipped, maintained and repaired in the physical world. A model that misreads a sentence can be corrected with a retrain; an arm that misjudges a grip can damage property or injure a person nearby.

That mismatch between software speed and hardware reality is where Karpas is expected to focus. Nvidia has spent years selling the training and simulation stack that robotics developers rely on, from synthetic environments for testing to the inference chips that run models onboard. The company's bet is that a robot's 'ChatGPT moment' arrives when general-purpose models can control many body types without bespoke engineering for each one. Until that happens, most deployments stay inside warehouses, factories and labs, where the environment is predictable and the cost of failure is contained.

The stakes stretch well beyond one conference panel. Robotics startups have raised billions on the promise of general-purpose machines, and investors are watching for signs that the timeline is shortening rather than stretching. Humanoid developers in particular have demoed impressive footage while shipping very few units to paying customers. If Karpas can articulate a concrete path from simulation to reliable field work, it gives the sector a clearer benchmark than another viral clip.

What to watch coming out of Disrupt is whether Karpas names specific bottlenecks, such as data collection for manipulation tasks, battery life, or the cost of safety-certified components. Naming them matters more than repeating that robots are coming. The open question is whether the industry's ChatGPT moment is a matter of a few years or a decade, and whether the companies that survive will be the ones building bodies or the ones building the brains.

TechCrunch Disrupt 2026 will host the session, and attendees who register before September 25 can save up to $200 on their pass.